1 Introduction
Cancer has a high morbidity and mortality and is one of the health concerns of
widespread concern worldwide. In recent years, several cancers with high incidence
rates are lung cancer, breast cancer, colon cancer, and stomach cancer. Among them,
the incidence of lung cancer in the population is greater than 10%, the mortality rate is
higher than the incidence rate, and the mortality rate exceeds 18%. According to the
statistics of lung cancer, the incidence of lung cancer in males ranks first, and the
mortality rate also ranks first. The incidence of lung cancer in women ranks second,
and mortality and morbidity rank second. Therefore, how to prevent and diagnose lung
cancer is one of the issues of widespread concern worldwide [1]. In recent years,
computer-aided diagnosis (CAD) can make good use of these unused CT images to
reduce the probability of missed diagnosis and misdiagnosis. The CAD system can use
the massive CT image of the lungs in the hospital to extract the features of the image
and then train the model. The CAD system can use the advantages of computers to
detect and diagnose CT images of lung nodules [2].
2 Materials and Methods
2.1 LIDC-IDRI
The source of lung CT data used in this article is from the National Cancer Institute
(NCI) LIDC-IDRI (The Lung Image Database Consortium) database. LIDC-IDRI has a
total of 1018 research examples. For each of the images in the example, there are four
experienced doctors who diagnose and label the two phases. In the first stage, four
doctors independently diagnosed and marked the lesions. The labels are divided into
three categories: nodules larger than or equal to 3 mm in diameter, nodules smaller than
3 mm, and pulmonary nodules greater than or equal to 3 mm. In the second phase, each
doctor needs to review the annotations given by several other doctors and give their
own diagnosis. The diagnosis of these two stages can minimize the misdiagnosis
caused by irresistible factors [3].
The image file format in the database is an image file. DICOM is one of the most
common format for medical images, which is shown in Fig.1.
Before using the algorithm to extract the features of the lung CT image, it is
necessary to analyze the lung CT image, determine the location of the pulmonary
nodule, segment the lung nodule image, and perform corresponding preprocessing on
the lung nodule image. Since the focus of this study is on computer-aided diagnosis of
pulmonary nodules rather than computerized detection of pulmonary nodules, this
paper uses the lung nodule contour coordinates given by the LIDC-IDRI database to
segment the lung nodule data [4] which is shown in Fig. 2.
498
Y. Li et al.
Cancer has a high morbidity and mortality and is one of the health concerns of
widespread concern worldwide. In recent years, several cancers with high incidence
rates are lung cancer, breast cancer, colon cancer, and stomach cancer. Among them,
the incidence of lung cancer in the population is greater than 10%, the mortality rate is
higher than the incidence rate, and the mortality rate exceeds 18%. According to the
statistics of lung cancer, the incidence of lung cancer in males ranks first, and the
mortality rate also ranks first. The incidence of lung cancer in women ranks second,
and mortality and morbidity rank second. Therefore, how to prevent and diagnose lung
cancer is one of the issues of widespread concern worldwide [1]. In recent years,
computer-aided diagnosis (CAD) can make good use of these unused CT images to
reduce the probability of missed diagnosis and misdiagnosis. The CAD system can use
the massive CT image of the lungs in the hospital to extract the features of the image
and then train the model. The CAD system can use the advantages of computers to
detect and diagnose CT images of lung nodules [2].
2 Materials and Methods
2.1 LIDC-IDRI
The source of lung CT data used in this article is from the National Cancer Institute
(NCI) LIDC-IDRI (The Lung Image Database Consortium) database. LIDC-IDRI has a
total of 1018 research examples. For each of the images in the example, there are four
experienced doctors who diagnose and label the two phases. In the first stage, four
doctors independently diagnosed and marked the lesions. The labels are divided into
three categories: nodules larger than or equal to 3 mm in diameter, nodules smaller than
3 mm, and pulmonary nodules greater than or equal to 3 mm. In the second phase, each
doctor needs to review the annotations given by several other doctors and give their
own diagnosis. The diagnosis of these two stages can minimize the misdiagnosis
caused by irresistible factors [3].
The image file format in the database is an image file. DICOM is one of the most
common format for medical images, which is shown in Fig.1.
Before using the algorithm to extract the features of the lung CT image, it is
necessary to analyze the lung CT image, determine the location of the pulmonary
nodule, segment the lung nodule image, and perform corresponding preprocessing on
the lung nodule image. Since the focus of this study is on computer-aided diagnosis of
pulmonary nodules rather than computerized detection of pulmonary nodules, this
paper uses the lung nodule contour coordinates given by the LIDC-IDRI database to
segment the lung nodule data [4] which is shown in Fig. 2.
498
Y. Li et al.